This commit is contained in:
Nicholas Butzke
2026-08-02 02:20:02 -04:00
parent 61669627db
commit 922983429e
226 changed files with 34032 additions and 18521 deletions
+90 -53
View File
@@ -1,67 +1,104 @@
# Sound Design Plan
# Sound Design
The current sounds are procedurally generated sine/noise blends
(`generate_sounds.py`) — functional, but robotic. This doc is the path to
real sound design with zero budget.
Papaya Shooter's mix is **stylized impact, comfortable repetition**: graphic
and energetic enough to match the cel-shaded action, but deliberately free of
the clipped white-noise and narrow resonant tones that make a fast shooter
fatiguing.
## What's in place now
Character movement follows the same 90/10 rule as
`docs/CHARACTER_ART_DIRECTION.md`: about 90% clean motion/thrust/body and 10%
actuator, latch or servo. Mechanics explain a verb; they do not become a
constant texture over the character.
`AudioManager` (autoload, `globals/audio_manager.gd`) provides the plumbing
good sound design needs:
## Runtime mix
- **Bus layout**: `Master ← SFX ← {Weapons, Footsteps, UI}`, `Master ← Music`
— created automatically at boot, so volume sliders and effects (EQ,
compression, reverb zones) can target categories.
- **Variation playback**: drop `footstep_01.wav`, `footstep_02.wav`,
`footstep_03.wav` into `assets/sounds/` and `AudioManager.play_3d("footstep", pos)`
picks one at random.
- **Pitch randomization**: every playback is pitched ±6% by default — the
single cheapest trick to stop repeated sounds sounding fake.
- **Pooled 3D players**: `play_3d()` is safe to call every frame; no node
churn, no cut-off management.
`AudioManager` owns registration, aliases, variation, routing and pooled
playback. The hierarchy is:
```gdscript
# anywhere in gameplay code:
AudioManager.play_3d("ak47_fire", muzzle.global_position)
AudioManager.play_ui("hit_confirm")
```text
Master (hard limiter)
├── SFX
│ ├── Weapons (light parallel compression)
│ ├── Explosions (slower body compression)
│ ├── Impacts
│ ├── Footsteps
│ ├── Movement
│ ├── Ambience
│ ├── UI
│ └── Wind
└── Music
```
## Replacing the procedural sounds (priority order)
The limiter is only a safety net for coincident shots/explosions. Assets are
mastered below full scale, and category gain establishes the hierarchy before
the limiter is reached.
Sound is ~50% of "game feel" in an FPS. Replace in this order:
Legacy event ids such as `m4_fire`, `explosion`, `wind` and `ui_hover` resolve
to premium families in `globals/audio_manager.gd`. This preserves gameplay API
compatibility while keeping source filenames out of weapon and UI code.
1. **Weapon fire** (heard constantly): layered = body (low thump) + crack
(mid transient) + tail (room reflection). Sonniss GDC packs have complete
layered gunshots. One file per weapon is fine to start; add `_01.._03`
variations later.
2. **Hit confirm + kill confirm**: short, bright, satisfying (this is the
"touchdown" pillar in the design doc).
3. **Footsteps**: 4+ variations minimum, quieter than you think, on the
`Footsteps` bus so they can be ducked while firing.
4. **Movement verbs**: slide (cloth+concrete scrape), dash (air whoosh),
wall-run loop, jump/land (land intensity scaled by fall speed).
5. **Ambience**: a quiet room tone per map kills the "dead air" feeling.
6. **UI**: menu hover/click/equip from a Kenney UI pack.
## Asset construction
Where to get files: see [ASSET_SOURCES.md](ASSET_SOURCES.md#sounds) —
Sonniss GDC bundles first, Kenney CC0 packs second, Freesound (CC0 filter)
for gaps.
Run:
## Conventions
```powershell
python tools/build_audio_assets.py
python tools/audio_audit.py
```
- Files: `assets/sounds/<event>[_NN].wav` — 44.1 kHz, 16-bit, mono for 3D
positional sounds, stereo only for UI/music.
- Ids are auto-registered from filenames at boot (`ak47_fire.wav`
`"ak47_fire"`). Names containing `fire`/`reload` route to the Weapons bus,
`footstep` to Footsteps, everything else to SFX
(see `BUS_HINTS` in audio_manager.gd).
- Keep source loudness consistent: normalize weapon shots to about -6 dBFS
peak, footsteps/foley around -18 dBFS.
The builder is deterministic except for downloading the pinned CC0 menu track.
It:
## Later polish (cheap wins, in order)
- retains the transient of the CC0 firearm recordings;
- layers a low-passed body and a short, dark reflection tail;
- reduces the abrasive 35 kHz region and controls peaks;
- creates four concrete, metal, wood and glass footstep variations;
- constructs a 48.5-second speed-wind master plus softened scrape, wall-run and
grapple loops from recordings;
- constructs mono mechanical projectile loops for correct 3D localization;
- layers low relay/clunk UI hover/click/confirm/error/equip feedback;
- masters the CC0 menu loop to a conservative music level.
1. Distance low-pass on the Weapons bus (far gunfire sounds muffled).
2. A short duck (sidechain) on Footsteps/ambience when local weapon fires.
3. Reverb `AudioEffectReverb` per-map on SFX for indoor spaces.
4. Surface-dependent footsteps: raycast down, map material → sound id
(`footstep_metal`, `footstep_concrete`).
`tools/audio_audit.py` gates family coverage, 48 kHz format, mono/stereo intent,
clipping, DC offset, headroom, loop duration, seam continuity, spectral
flatness, high-frequency centroid and the explosion/firearm loudness hierarchy.
## Gameplay behavior
- Footsteps raycast the actual floor collider and use its
`acoustic_material`; brick and untagged structure map to concrete.
- Local and remote footsteps use the same surface families. Remote steps remain
positional and pass through material-aware occlusion.
- Slides switch recorded friction gestures if the floor material changes and
scale restrained pitch/gain with speed.
- Wall-run friction scales with speed and gets a restrained wood/metal
resonance shift.
- Speed wind is a narrowed, filtered field recording and fades in above 12 m/s
with a stronger velocity-pitch relationship. Its 48.5-second cycle prevents
a short gust cadence from becoming audible. There is no global noise bed.
- Jump and double-jump are compact jump-jet impulses. Dash is an equally brief
thrust-jet cue; clean air and low engine body dominate both.
- Grapple launch/latch/reel use distinct air, padded latch and restrained motor
cues. The reel stays below the character movement and never becomes a
continuous metallic foreground.
- Automatic weapon players keep eight overlapping voices; a new shot no longer
cuts the previous shot's body and tail.
- Explosion pitch and gain scale with blast radius, while reflections and
occlusion locate the source in the environment. Masters carry a measured
body advantage over firearms and a dark industrial debris layer.
- UI hover has a short cooldown so crossing dense menus does not become a buzz.
All UI feedback is built from softened relay/clunk gestures with 8 ms
transient ramps and a two-pole 2.4 kHz ceiling.
- Menu music fades in and is owned by the menu scene, so it stops cleanly when
a match begins.
## Source and naming conventions
Licenses and download pages are recorded in `assets/sounds/SOURCES.md`.
- Positional one-shots and projectile loops: 48 kHz, 16-bit mono WAV.
- Player-local movement beds: 48 kHz, 16-bit stereo WAV.
- Music: 48 kHz stereo Ogg Vorbis.
- Variations: `<event>_00.wav`, `<event>_01.wav`, etc.
- Runtime event families, bus gains and pitch ranges live in
`AudioManager.EVENT_TUNING`.